基于神经网络与粒子滤波的柔性臂控制方法研究

基于奇异摄动法将单连杆柔性臂系统分解为慢变、快变子系统,采用混合控制方法;设计了基于粒子滤波的神经网络控制器来线性化慢子系统,使其跟踪期望轨迹;采用粒子滤波训练神经网络克服了BP算法收敛速度慢、易陷入局部极小值的缺陷,及扩展卡尔曼滤波方法带来的模型线性化损失;对于快变系统采用最优控制方法;仿真结果表明:在神经网络训练误差收敛速度及精度方面,粒子滤波要比BP及卡尔曼滤波要好;组合控制方法能有效地抑制柔性臂弹性振动,轨迹跟踪迅速准确,精度方面也是前者最优。

Research on Controlling of Manipulator with Flexible Links Based on Neural Network and Particle Filter

The singular perturbation approach is used to decompose the manipulator with one flexible links into a slow and a fast system, which allows a composite control design for the original system.A neural network(NN)controller based on Particle Filter(PF)is designed to linearize the slow system and make the rigid motion track a desired trajectory.Using PF to train the neural networks can overcome the drawbacks of BP algorithm such as falling into local minima and slow convergence,and the linearized error of Extended Kalman Filter.An optimal controller is designed to stabilize the fast subsystem.Simulations results show that the error convergence speed and accuracy of NN when trained by PF is better then by BP and EKF;the composite controller is effective in motion tracking and vibration attenuation control, and the tracking accuracy of PF is best too.